Add training service #225
Open
Add training service #225
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
thodkatz
force-pushed
the
add-training-servicer
branch
from
bbf0243 to
ace0793
Compare
Codecov ReportAttention: Patch coverage is
Additional details and impacted files@@ Coverage Diff @@
## main #225 +/- ##
==========================================
+ Coverage 64.60% 65.52% +0.92%
==========================================
Files 40 44 +4
Lines 2195 2770 +575
==========================================
+ Hits 1418 1815 +397
- Misses 777 955 +178 ☔ View full report in Codecov by Sentry. |
thodkatz
force-pushed
the
add-training-servicer
branch
from
ace0793 to
cbb2619
Compare
thodkatz
commented
Dec 10, 2024
| @@ -1,7 +1,7 @@ | |||
| [pytest] | |||
| python_files = test_*.py | |||
| addopts = | |||
| --timeout 10 | |||
| --timeout 60 | |||
There was a problem hiding this comment.
The difference in creating threads, processes using a start method "spawn" instead of "fork" is quite significant, that led me to bump it, so the tests can pass for macos and windows platforms.
thodkatz
force-pushed
the
add-training-servicer
branch
3 times, most recently
from
e9fc4b0 to
f345f69
Compare
k-dominik
reviewed
Dec 11, 2024
Comment on lines
+187
to
+192
| def set_max_num_iterations(self, num_iterations: int): | ||
| raise NotImplementedError | ||
|
|
||
| def update_dataset(self): | ||
| raise NotImplementedError | ||
|
|
There was a problem hiding this comment.
why does the biomodel supervisor need those?
There was a problem hiding this comment.
I was hesitant to remove them as a reference of what the old design was intented to do. But I wasn't sure what to do with these. I could move them to the trainer one maybe
thodkatz
force-pushed
the
add-training-servicer
branch
3 times, most recently
from
750ee88 to
d432e25
Compare
- Supported operations: start, resume, pause, shutdown - pytorch-3dunet package is used as the framework to create the models
… failed I caught an edge case, where events are blocked, because we have exited the training, and the tasks of the queue would remain unprocessed.
Creating and closing processes and threads can be quite time consuming resulting to test timeouts if the tests performs a lot of actions.
Applying monkeypatch to a parent process won't propagated to a child process if the start method is spawn (macos) instead of fork (linux)
- To fix test on windows, convert label data to float64
The should stop callbacks are boolean, so we need to aggregate their return value. Previously the return value wasn't taken into account, and the callbacks were returning none
The enum is used for validation check before triggering one of them. Previously I was checking if the queue was alive, but that won't be enough, for example if you want to perform resume, while you are resumed, the queue is operational, but the action shouldn't be valid.
thodkatz
force-pushed
the
add-training-servicer
branch
6 times, most recently
from
b9153b1 to
a088627
Compare
thodkatz
force-pushed
the
add-training-servicer
branch
3 times, most recently
from
09fcb85 to
4a5ed85
Compare
thodkatz
force-pushed
the
add-training-servicer
branch
2 times, most recently
from
52fb50f to
752cb59
Compare
Move NamedInt and Tensor proto to a separate file so training proto can use as well
- The inference servicer had a procedure to list the available devices. This is needed or the training servicer as well. So list devices is decoupled to be shared.
thodkatz
force-pushed
the
add-training-servicer
branch
from
752cb59 to
27b3923
Compare
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
An initial implementation of adding neural network training in tiktorch :P
It requires:
The package
pytorch-3dunetis used as our framework to configure the models.Supported workflows:
The above are demonstrated with tests. Please check them first :)
For thread synchronization we rely on a thread-safe priority queue (todo: explain the scheme with the concept of ICommand, and the error handling).
Manual testing
I have created a setup to test the server with actual data. You can find more here: https://github.com/thodkatz/ilastik-playground